Author: Uri Blass
Date: 02:48:01 04/08/04
Go up one level in this thread
On April 08, 2004 at 05:23:33, Tord Romstad wrote: >On April 08, 2004 at 01:02:42, Mridul Muralidharan wrote: > >>Hi, >> >> Importance of endgame knowledge has been time and again discussed here. >>But it is quiet frustrating for a chess patzer like me to implement this in the >>engine - much more if I attempt to do this effectively. > >I agree entirely. Adding endgame knowledge is very difficult, especially if you >want to >keep the code general rather than writing huge amounts of code for dozens of >different >special cases. I think that it is dependent on the knowledge that you have. based on the game it seems that it not very difficult in the relevant case to get significant improvement for Mridul's program. And if you give up and start writing code for all sorts of >special cases, you >face the ugly problem of evaluation discontinuities when pieces are exchanged. It dependent on what you write. You can decide to write code only for clear cases when you can get win,draw,loss by evaluation and in this case I do not see the problem of discontinuities. > >Not only the eval, but also the search is difficult in the endgame. All the >well-known >selective search techniques seem to fail miserably in the endgame. I found that for me verified null move pruning helped in the endgame and today I use it only in the endgame when normal null move pruning is used in the middle game. I think that the right selective search can help more in the endgame then in the middle game. The point is that programmers usually care more about the middle game because the middle game is more important. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.